Skip to main content
PromptQuorum
Home/Power Local LLM/ClawRouter Review: The Agent-Native LLM Router (2026)
Overview & Reference

ClawRouter Review: The Agent-Native LLM Router (2026)

Β·11 min readΒ·By Hans Kuepper Β· Founder of PromptQuorum, multi-model AI dispatch tool Β· PromptQuorum

ClawRouter is a free, open-source (MIT) LLM router that makes local, sub-millisecond routing decisions across 78 chat models from nine providers, then forwards each request to the model its scorer picks β€” paid either with USDC micropayments an autonomous agent can sign itself, or with an ordinary credit card for human users. Built by BlockRunAI, it runs as a local Node.js proxy (npx @blockrun/clawrouter) on port 8402, plugs into OpenClaw as a first-party plugin, and works with any OpenAI-compatible client such as continue.dev or Cursor.

ClawRouter (github.com/BlockRunAI/ClawRouter) is a free, open-source (MIT) LLM router built by BlockRunAI that makes local, sub-millisecond routing decisions across dozens of chat models, then forwards each request to whichever model wins that decision. It is the companion CLI/proxy to the BlockRun gateway, and its distinguishing feature is that it can be paid entirely by an autonomous agent β€” a locally generated wallet signs USDC micropayments over the x402 protocol, with no account, API key, or credit card required β€” while still accepting a normal credit card for human users. This review covers what it does, how its routing and payment model actually work, and who it fits.

Key Takeaways

  • ClawRouter (github.com/BlockRunAI/ClawRouter) is a free, open-source, self-hosted LLM router built by BlockRunAI
  • MIT licensed, confirmed via the GitHub repository's license field
  • Routing decisions happen 100% locally, in under 1ms, using the open-source @blockrun/router-core scoring engine, before the winning request is forwarded to an external model provider
  • Six open-weight models are free forever with no signup, no API key, and no wallet required
  • 78 chat models across nine providers (OpenAI, Anthropic, Google, xAI, DeepSeek, and others) behind one local endpoint
  • Two payment paths: USDC micropayments over the x402 protocol on Solana or Base (an autonomous agent can pay for itself), or a credit-card-funded account balance at user.blockrun.ai
  • Over 6,600 GitHub stars and 650+ forks as of this review
  • Distributed as the @blockrun/clawrouter npm package; also installs as a plugin for the OpenClaw coding agent

πŸ“ In One Sentence

ClawRouter is a free, open-source (MIT) LLM router built by BlockRunAI that scores each request across 15 dimensions in under 1ms entirely on your own machine, then routes it to the cheapest capable model among 78 options, paid either through agent-signed USDC micropayments or a human credit card.

πŸ’¬ In Plain Terms

ClawRouter is a small program you run on your own computer or server that decides, for every AI request, which AI model should answer it β€” a cheap one for a simple question, a more capable one for a hard task β€” without you picking manually. Six models cost nothing at all. For the paid ones, an autonomous agent can pay by signing a tiny cryptocurrency payment itself, or a person can just add a credit card, no wallet required.

πŸ“ŒNote: This review is the deep-dive companion to ClawRouter's entry in the Local LLM Software Directory β€” see that page for how ClawRouter compares at a glance to dozens of other local AI tools. It is based on ClawRouter's own README, GitHub repository, and release history, not hands-on benchmarking by PromptQuorum.

What Is ClawRouter?

ClawRouter is a self-hosted LLM router that picks the cheapest model capable of handling each request, then forwards that request to an external provider β€” it does not run any model inference itself. Its own GitHub description calls it "the agent-native LLM router for autonomous agents," and it is built and maintained by BlockRunAI, the same organization behind the BlockRun payment gateway that settles the paid requests it routes.

  • Product type: a local Node.js proxy and CLI, distributed as the @blockrun/clawrouter npm package β€” not a GUI desktop app, though an optional macOS Electron control-plane app exists alongside it
  • Creator: BlockRunAI
  • Repository: github.com/BlockRunAI/ClawRouter, created February 3, 2026
  • License: MIT, confirmed via the GitHub repository's license metadata
  • Locality: routing decisions are made 100% locally in under 1ms; the request itself is then forwarded to an external model provider over the BlockRun gateway β€” ClawRouter does not run inference on your own hardware
  • Scale: over 6,600 GitHub stars, 650+ forks, as of this review

Project History and Version Milestones

The ClawRouter GitHub repository was created on February 3, 2026, and the project has shipped frequent point releases since, including a desktop control-plane app for macOS. GitHub's release API shows dense, near-daily versioning in its most recent activity window.

  1. 1
    Repository created β€” February 3, 2026
    Why it matters: Marks the start of ClawRouter's public GitHub history, per the repository's own creation timestamp.
  2. 2
    ClawRouter Desktop 0.1.3 Preview β€” September 8, 2026
    Why it matters: A preview release of the optional macOS Electron control-plane app, with a UI refresh and status fixes.
  3. 3
    v0.12.276 β€” September 7, 2026: Gemini 3.8 Flash catalog fix
    Why it matters: Fixed a model that was already routable but missing from the public catalog listing, per the release notes.
  4. 4
    v0.12.277 β€” September 7, 2026: Opt-in TWZRD AutoGate
    Why it matters: Introduced an opt-in routing feature (TWZRD AutoGate) alongside an availability fix, per the release notes.
  5. 5
    v0.12.278 β€” September 16, 2026: TWZRD AutoGate removed
    Why it matters: Removed the AutoGate feature introduced one release earlier, per the release notes β€” worth noting for anyone who had opted in.
  6. 6
    v0.12.279 β€” September 16, 2026: Desktop control-plane refresh
    Why it matters: Added a restart-free model-chain switch and a Hono framework patch, per the release notes β€” the most recent tagged release as of this review.

What Does ClawRouter Actually Do?

ClawRouter's feature set centers on making a routing decision locally and settling payment for the result, without requiring the human-oriented signup flow most LLM gateways expect. Here is what each part actually does, per ClawRouter's own GitHub README.

  • Local, sub-millisecond routing β€” a 15-dimension weighted scorer, running entirely on your own machine via the open-source @blockrun/router-core engine, classifies each request's difficulty tier and task shape, then filters and ranks candidate models before picking a winner in under 1ms
  • Routing profiles β€” /model free (free models only), /model auto (balanced, the default), /model eco (cheapest viable model), and /model premium (best quality), switchable per conversation
  • Agent-native payment β€” a wallet is generated locally on first run; an autonomous agent signs a USDC micropayment over the x402 protocol on Solana or Base for each paid request, with no account, API key, or credit card involved
  • Human-friendly payment β€” sign up at user.blockrun.ai, add a credit card, and mint an API key (clawrouter login brk_live_…); ClawRouter then uses that key instead of the wallet, billing a card-funded account balance
  • Free tier β€” six open-weight models (including a long-context reasoner and two fast coder-tuned models per the README) are free forever with no signup, no key, and no wallet
  • Fallback chains β€” every routing tier carries a benchmark-ordered fallback list, so a provider timeout or server error triggers a retry against the next-ranked model rather than failing the request outright
  • OpenAI-compatible proxy β€” runs a local HTTP proxy on port 8402 that any OpenAI-compatible client (continue.dev, Cursor, custom OpenAI SDK code) can point at directly
  • OpenClaw plugin β€” installs as a first-party plugin for the OpenClaw coding agent, registering blockrun/auto as a selectable model
  • Desktop control plane (macOS) β€” an optional Electron app that provides one-click connections for OpenClaw, Codex, Hermes, and other agents, plus model pricing, wallet balances, and chain switching between Solana and Base

Usage Examples: Three Ways to Use ClawRouter

These are concrete workflows built from ClawRouter's documented features above β€” not hypothetical use cases.

Platform, Pricing, and Licensing

Platform

What ClawRouter states:
A local Node.js CLI/proxy (any OS with Node.js and npm/npx); optional macOS Electron desktop control-plane app.

Cost

What ClawRouter states:
Free and open source. Six models free forever; paid models billed via USDC micropayments or a card-funded balance.

Licensing

What ClawRouter states:
MIT, confirmed via the GitHub repository's license metadata.

Install method

What ClawRouter states:
Per the README: npx @blockrun/clawrouter, or npm install -g @blockrun/clawrouter then clawrouter setup.

Verify the current recommended install method on github.com/BlockRunAI/ClawRouter before running either command, since install instructions can change between releases.

ClawRouter Pricing: Free Models, USDC, and Credit Cards

ClawRouter itself is free and open source; what costs money is the paid model usage it routes to, billed one of two ways. There is no subscription and no seat-based fee for ClawRouter as software β€” the GitHub LICENSE is a standard MIT license with no feature gated behind payment.

  • Free tier: six open-weight models, free forever, no signup, no API key, no wallet required
  • USDC path: a locally generated wallet pays per-request over the x402 protocol on Solana (default) or Base; a flat $0.001 settlement fee per transaction covers on-chain gas, non-custodial
  • Credit-card path: sign up at user.blockrun.ai, top up with a card, mint an API key; a 5.5% + $0.30 fee applies on the top-up itself, then requests draw down that balance at list price with no additional per-call fee
  • No ClawRouter-specific markup on model pricing beyond the payment-path fees above, per the README's own comparison table
  • Running ClawRouter locally costs nothing beyond the model usage you actually route β€” no ClawRouter subscription exists

ClawRouter vs. LiteLLM

ClawRouter and LiteLLM are both open-source LLM routing/gateway tools that run locally, but they differ in how routing decisions get made and how requests get paid for. LiteLLM is a widely used, bring-your-own-API-keys gateway with manual model selection; ClawRouter adds an automated local scoring engine and a payment layer that does not require a human to hold API keys for every provider.

Aspect
ClawRouter
LiteLLM
RoutingAutomated 15-dimension local scorer (router-core)Manual model selection per request
Auth/paymentWallet signature (USDC/x402) or card-funded API keyBring your own API keys per provider
Free tierSix models free forever, no signupFree software, but you pay each provider directly
Runs locallyYes, local proxy on port 8402Yes, self-hosted gateway/proxy
LicenseMITSee the dedicated LiteLLM review for current license terms

This comparison reflects each project's own documentation, not an independent PromptQuorum benchmark. See the full LiteLLM review for details on that tool specifically before choosing between them.

Who Should Use ClawRouter?

Whether ClawRouter fits depends on whether you want automated, local model routing plus a payment path that does not require a human to hold API keys for every provider.

Competitors and Alternatives

ClawRouter sits in the routers-and-gateways segment of local AI tooling, alongside other self-hosted LLM proxies. Its main differentiator is the combination of fully local routing decisions with an agent-native, wallet-based payment path β€” most routers in this segment assume a human holding API keys.

Tool
Best known for
Link
LiteLLMWidely used open-source LLM gateway/proxy with manual routing and bring-your-own-keysLiteLLM review
PlanoLLM routing/gateway tool covered separately in this clusterPlano review

This list reflects other routers/gateways covered in PromptQuorum's Local LLM Software Directory, not an independent ranking β€” verify each tool's current pricing and feature set before choosing. See the Local LLM Software Directory for the full catalog.

Common Mistakes When Evaluating ClawRouter

Most confusion about ClawRouter comes from assuming it requires cryptocurrency, or from conflating its local routing decision with local model inference.

Frequently Asked Questions

What is ClawRouter?

ClawRouter (github.com/BlockRunAI/ClawRouter) is a free, open-source (MIT) LLM router built by BlockRunAI. It scores each request locally in under 1ms and routes it to the cheapest capable model among 78 options, then settles payment via USDC micropayments or a credit-card-funded balance.

Is ClawRouter free?

The software itself is free and open source with no subscription. Six open-weight models are free forever to use through it. Paid frontier models cost whatever that model's list price is, billed via USDC or a card-funded balance β€” ClawRouter does not add its own markup.

How do I install ClawRouter?

Per the official README, run npx @blockrun/clawrouter to try it immediately, or npm install -g @blockrun/clawrouter followed by clawrouter setup for a persistent install. Verify current instructions on github.com/BlockRunAI/ClawRouter before installing.

Does ClawRouter require cryptocurrency?

No. A locally generated wallet and USDC micropayments are the default path for autonomous agents, but human users can sign up at user.blockrun.ai, add a credit card, and run clawrouter login to bill a card-funded balance instead.

Does ClawRouter run AI models locally?

No. Only the routing decision β€” which model should handle a given request β€” runs locally, in under 1ms. The actual model response is served by an external provider (OpenAI, Anthropic, Google, DeepSeek, and others) behind the BlockRun gateway.

What license does ClawRouter use?

MIT, confirmed via the GitHub repository's license metadata.

Which AI models does ClawRouter support?

78 chat models across nine providers as of this review, per ClawRouter's own model catalog at blockrun.ai/models β€” including OpenAI, Anthropic, Google, xAI, and DeepSeek models, plus six free open-weight models.

How does ClawRouter integrate with OpenClaw?

ClawRouter installs as a first-party OpenClaw plugin (plugin id blockrun-clawrouter), registering blockrun/auto as a selectable model. A one-shot install script or clawrouter setup after npm install -g completes the OpenClaw integration.

Can I use ClawRouter with continue.dev or Cursor?

Yes. ClawRouter runs a local OpenAI-compatible proxy on port 8402; both continue.dev and Cursor (and any OpenAI SDK-based tool) can be pointed at that local endpoint directly.

Who built ClawRouter?

BlockRunAI, the organization behind the BlockRunAI/ClawRouter GitHub repository and the BlockRun payment gateway that settles ClawRouter's paid requests.

Sources

← Back to Power Local LLM